aditi gunna

STORE TO SCREEN: DESIGNING A VOICE AI ASSISTANT

[INTERACTION DESIGN]
[UI/UX DESIGN]



How might we make a voice interface feel like talking to someone in a store? 
For a major telecom client, I worked on a voice AI interface that lets customers complete tasks like purchasing a device or troubleshooting an issue, entirely through conversation. The concept was built as an MVP to demo this technology to the client and is currently in development.


Project
For a major telecom client at Cognizant Time Frame 
2 Weeks, October-November 2023  My Role
Interaction and visual design, prototyping. 


Design Solution 

 In a physical store, an employee listens to what someone needs, thinks through the best option in real time, and brings solutions directly to them. The goal of this project was to recreate that same sense of attentiveness entirely through voice.

What we developed was a voice assistant that launches as an overlay directly from the site. Customers interact almost entirely by speaking; the assistant listens, responds, and surfaces relevant information on screen as the conversation unfolds. We designed and prototyped two workflows to test the concept: a sales journey, where the assistant helps a customer choose and purchase a device, and a support journey, where it helps troubleshoot an issue with an existing one.

Voice as the primary interactionNearly all interaction happens through natural conversation rather than clicking through menus. 

An agentic AI system listens to what the customer says, interprets intent, and determines the next step, including taking action directly when appropriate.




Two states: listening and speaking
A glowing gradient border around the whole panel signals listening; a glowing gradient aura around the assistant's avatar signals speaking. Clicking is reserved for controlling the session itself, like starting, muting, or pausing.

 




A visual layer that follows the conversation
The visual UI updates dynamically based on what's being discussed. If the assistant recommends a few device options, those appear on screen, the way an employee might lay a few choices in front of you.




Confirming every choice, twice
Any selection, or any reference the assistant makes to something on screen, is highlighted visually. Decisions are reaffirmed both verbally and visually before moving forward.

Design Process

Most existing UX literature, including foundational texts like Steve Krug's Don't Make Me Think, treats the web page as the core unit of interaction: information organized into a "page," navigated by scanning and clicking, much like reading a book. A voice-first interface breaks that metaphor entirely. There is no page to scan, no menu to scroll through. The "page" becomes a conversation: linear, responsive, and impossible to skim ahead.

Removing the page also meant removing most visual feedback. In a typical app, a spinner or a highlighted button tells you something is happening; voice doesn't have that built in. We borrowed a parallel from video calls, which are, in a sense, already a kind of digital conversation. That comparison surfaced two essential feedback states: the customer needs to know when the assistant is listening, and when it's speaking. We also limited click-based interaction by design. That decision raised the stakes on clarity: if a customer can't click their way out of confusion, every choice the system makes on their behalf needs to be visible and easy to confirm.

Outcome

The result was an MVP designed to demo the concept to the client, built to be versatile enough to layer onto an existing interface rather than exist as a standalone product. It's currently in development.
prev